Computer systems today are powerful, but are rendered limited in their ability to be divided into modular components due to a variety of technical limitations of today's PCI bus technology. And in their ability to adapt to changing computing environments. The PCI bus is pervasive in the industry, but as a parallel data bus is not easily extended over any distance or bridged to other remote PCI based devices due to loading and physical constraints, most notably the inability to extend the PCI bus more than a few inches. Full bridges are known, such as used in traditional laptop computer/docking stations. However, separating the laptop computer from the docking station a significant distance has not been possible. Moreover, the processing power of computer systems has been resident within the traditional computer used by the user because the microprocessor traditionally is directly connected to and resident on the PCI motherboard. Thus, upgrading processing power usually meant significant costs and/or replacing the computer or computer system.
PCI
The PCI bus is primarily a wide multiplexed address and data bus that provides support for everything from a single data word for every address to very long bursts of data words for a single address, with the implication being that burst data is intended for sequential addresses. Clearly the highest performance of the PCI bus comes from the bursts of data, however most PCI devices require reasonable performance for even the smallest single data word operations. Many PCI devices utilize only the single data mode for their transfers. In addition, starting with the implementation of the PCI 2.1 version of the specification, there has been at least pseudo isochronous behavior demanded from the bus placing limits on an individual device's utilization of the bus, thus virtually guaranteeing every device gets a dedicated segment of time on a very regular interval and within a relatively short time period. The fundamental reason behind such operation of the PCI bus is to enable such things as real time audio and video data streams to be mixed with other operations on the bus without introducing major conflicts or interruption of data output. Imagine spoken words being broken into small unconnected pieces and you get the picture. Prior to PCI 2.1 these artifacts could and did occur because devices could get on the bus and hold it for indefinite periods of time. Before modification of the spec for version 2.1, there really was no way to guarantee performance of devices on the bus, or to guarantee time slot intervals when devices would get on the bus. Purists may argue that PCI is still theoretically not an isochronous bus, but as in most things in PC engineering, it is close enough.
Traditional High Speed Serial
Typical high speed serial bus operation on the other hand allows the possibility of all sizes of data transfers across the bus like PCI, but it certainly favors the very long bursts of data unlike PCI. The typical operation of a serial bus includes an extensive header of information for every data transaction on the bus much like Ethernet, which requires on the order of 68 bytes of header of information for every data transaction regardless of length. In other words, every data transaction on Ethernet would have to include 68 bytes of data along with the header information just to approach 50% utilization of the bus. As it turns out Ethernet also requires some guaranteed dead time between operations to “mostly” prevent collisions from other Ethernet devices on the widely disperse bus, and that dead time further reduces the average performance.
The typical protocol for a serial bus is much the same as Ethernet with often much longer header information. Virtually all existing serial bus protocol implementations are very general and every block of data comes with everything needed to completely identify it. FiberChannel (FC) has such a robust protocol that virtually all other serial protocols can be transmitted across FC completely embedded within the FC protocol, sort of like including the complete family history along with object size, physical location within the room, room measurements, room number, street address, city, zip code, country, planet, galaxy, universe, . . . etc. and of course all the same information about the destination location as well, even if all you want to do is move the object to the other side of the same room. Small transfers across many of these protocols, while possible, are extremely expensive from a bandwidth point of view and impractical in a bus applications where small transfers are common and would be disproportionally burdened with more high overhead than actual data transfer. Of course the possibility of isochronous operation on the more general serial bus is not very reasonable.
Recreating High Speed Serial for PCI
In creating the proprietary Split-Bridge™ technology, Mobility electronics of Phoenix Ariz., the present applicant, actually had to go back to the drawing board and design a far simpler serial protocol to allow a marriage to the PCI bus, because none of the existing implementations could coexist without substantial loss of performance. For a detailed discussion of Applicant's proprietary Split-Bridge™ technology, cross reference is made to Applicant's co-pending commonly assigned patent applications identified as Ser. No. 09/130,057 and 09/130,058 both filed Aug. 6, 1998, the teachings of each incorporated herein by reference. The Split-Bridge™ technology approach is essentially custom fit for PCI and very extensible to all the other peripheral bus protocols under discussion like PCIx, and LDT™ set forth by AMD Corporation. LDT requires a clock link in addition to its data links, and is intended primarily as a motherboard application, wherein Split-Bridge™ technology is primarily intended to enable remote bus recreation. As the speeds of motherboard buses continue to grow faster, Split-Bridge™ can be readily adapted to support these by increasing the serial bus speed and adding multiple pipes. Split-Bridge™ technology fundamentals are a natural for extending anything that exists within a computer. It basically uses a single-byte of overhead for 32 bits of data and address—actually less when you consider that byte enables, which are not really “overhead”, are included as well.
Armed with the far simpler protocol, all of the attributes of the PCI bus are preserved and made transparent across a high speed serial link at much higher effective bandwidth than any existing serial protocol. The net result is the liberation of a widely used general purpose bus, and the new found ability to separate what were previously considered fundamental inseparable parts of a computer into separate locations. When the most technical reviewers grasp the magnitude of the invention, then the wheels start to turn and the discussions that follow open up a new wealth of opportunities. It now becomes reasonable to explore some of the old fundamentals, like peer-to-peer communication between computers that has been part of the basic PCI specification from the beginning, but never really feasible because of the physical limits of the bus prior to Split-Bridge™ technology. The simplified single-byte overhead also enables very efficient high speed communication between two computers and could easily be extended beyond PCI.
The proprietary Split-Bridge™ technology is clearly not “just another high speed link” and distinguishing features that make it different represent novel approaches to solving some long troublesome system architecture issues.
First of all is the splitting of a PCI bridge into two separate and distinct pieces. Conceptually, a PCI bridge was never intended to be resident in two separate modules or chips and no mechanism existed to allow the sharing of setup information across two separate and distinct devices. A PCI bridge requires a number of programmable registers that supply information to both ports of a typical device. For the purpose of the following discussion, the two ports are defined into a north and south segment of the complete bridge.
The north segment is typically the configuration port of choice and the south side merely takes the information from the registers on the north side and operates accordingly. The problem exists when the north and south portions are physically and spatially separated and none of the register information is available to the south side because all the registers are in the north chip. A typical system solution conceived by the applicant prior to the invention of Split-Bridge™ technology would have been to merely create a separate set of registers in the south chip for configuration of that port. However, merely creating a separate set of registers in the south port would still leave the set up of those registers to the initialization code of the operating system and hence would have required a change to the system software.
Split-Bridge™ technology, on the other hand, chose to make the physical splitting of the bridge into two separate and spaced devices “transparent” to the system software (in other words, no knowledge to the system software that two devices were in fact behaving as one bridge chip). In order to make the operations transparent, all accesses to the configuration space were encoded, serialized, and “echoed” across the serial link to a second set of relevant registers in the south side. Such transparent echo between halves of a PCI bridge or any other bus bridge is an innovation that significantly enhances the operation of the technology.
Secondly, the actual protocol in the Split-Bridge™ technology is quite unique and different from the typical state of the art for serial bus operations. Typically transfers are “packetized” into block transfers of variable length. The problem as it relates to PCI is that the complete length of a given transfer must be known before a transfer can start so the proper packet header may be sent.
Earlier attempts to accomplish anything similar to Split-Bridge™ technology failed because the PCI bus does not inherently know from one transaction to the next when, or if, a transfer will end or how long a block or burst of information will take. In essence the protocol for the parallel PCI bus (and all other parallel, and or real time busses for that matter) is incompatible with existing protocols for serial buses.
An innovative solution to the problem was to invent a protocol for the serial bus that more or less mimics the protocol on the PCI. With such an invention it is now possible to substantially improve the performance and real time operation here to for not possible with any existing serial bus protocol.
The 8 bit to 10 bit encoding of the data on the bus is not new, but follows existing published works. However, the direct sending of 32 bits of information along with the 4 bits of control or byte enables, along with an additional 4 bits of extension represents a 40 bit for every 36 bits of existing PCI data, address, and control or a flat 10% overhead regardless of the transfer size or duration, and this approach is new and revolutionary. Extending the 4 bit extension to 12 or more bits and including other functionality such as error correction or retransmit functionality is also within the scope of the Split-Bridge™ technology.
New Applications of the Split-Bridge™ Technology
Basic Split-Bridge™ technology was created for the purpose of allowing a low cost, high speed serial data communications between a parallel system bus and remote devices. By taking advantage of the standard and pervasive nature of the PCI bus in many other applications in computing, dramatic improvements in the price performance for other machines is realized. The present invention comprises a revolutionary application rendered possible due to the attributes of applicant's proprietary Split-Bridge™ technology.